Hand Component Decomposition for the Hand Gesture Recognition Based on FingerPaint Dataset

In the human-machine interaction system, hand component decomposing is important to recogtlize the human gesture. This paper proposes a method to decompose the hand component for the hand gesture recognition from human body image of FingerPaint dataset which is Microsoft research open data. We choose 36,750 randomly images for training and choose the remaining 15,750 images for the testing from the FingerPaint dataset. We conducted the PFACA(Proportion of frames with average classification accuracy) for the accuracy of the hand component(thumb, index finger, middle finger, ring finger, pinky, palm, wrist). In the results of five times repeated experiments that we showed maximum of 0.9849042 and minimum of 0.949042 at a frame of more than 0.2 of Epsilon.